Computational Representations of Character Significance in Novels

TL;DR

This work introduces a six-component model of character significance (N, C, I, A, DC, DN) to capture how characters occupy narrative space beyond mere presence, including narration and contextual discussions. It couples expert-annotated ground truth with automated labeling (BookNLP and span-level LLMs) and builds three networks—co-occurrence, dialogue, and directed discussion—to study character centrality at both component and network levels. Applying the pipeline to Pride and Prejudice, Jane Eyre, and a corpus of 64 nineteenth-century novels, the study demonstrates that the directed discussion network reveals gendered patterns and offers a nuanced perspective on centrality that extends beyond Woloch's 'one vs the many.' The results establish a scalable computational framework for literary analysis, releasing datasets and tools that enable corpus-wide inferences about character prominence, narrative space, and gender dynamics in classic fiction.

Abstract

Characters in novels have typically been modeled based on their presence in scenes in narrative, considering aspects like their actions, named mentions, and dialogue. This conception of character places significant emphasis on the main character who is present in the most scenes. In this work, we instead adopt a framing developed from a new literary theory proposing a six-component structural model of character. This model enables a comprehensive approach to character that accounts for the narrator-character distinction and includes a component neglected by prior methods, discussion by other characters. We compare general-purpose LLMs with task-specific transformers for operationalizing this model of character on major 19th-century British realist novels. Our methods yield both component-level and graph representations of character discussion. We then demonstrate that these representations allow us to approach literary questions at scale from a new computational lens. Specifically, we explore Woloch's classic "the one vs the many" theory of character centrality and the gendered dynamics of character discussion.

Figures (8)

• Figure 1: Plots showing component scores from our span-level methods for top characters in Pride and Prejudice. The left plot shows total component score for all tags across the top 10 characters and the right plot shows a breakdown of tags by component for the protagonist, Elizabeth.
• Figure 2: Character networks for Pride & Prejudice illustrating the three edge representations chosen.
• Figure 3: Poincaré disk embedding of the Middlemarch co-occurrence network. Radial position encodes character prominence; angular position reflects community structure. The embedding recovers the novel's multi-plot organization.
• Figure 4: Plots showing the total component scores for our span-level methods on the bottom 20 characters in Pride and Prejudice and Jane Eyre. We see much greater variability in this set of characters than for the top characters, in part because the total counts are much lower. For Jane Eyre, the BookNLP approach uses the manual correction of the co-reference issue for Jane.
• Figure 5: Plots showing component scores from our span-level methods for top characters in Jane Eyre. The left plot shows total component score for all tags across the top 10 characters and the right plot shows a breakdown of tags by component for the protagonist, Jane.